Comment 67 for bug 249340

I'm a bit surprised that noone else seems to be considering how to *fix* this problem. All I'm hearing are work-arounds for users, which is not really adequate.

It's pretty clear that there's a kernel bug -- nothing else explains the "unkillable processes" problem (which I can personally vouch for using linux-image-2.6.22-15-generic). The only question is whether the kernel bug is new, or whether something changed in localedef so that it now triggers a bug that already existed.

I earlier mentioned that I suspected that this was related to kernel bug #231746, which indicates that the problem is triggered by trying to write a zero-length segment to a file -- either a malformed descriptor for "writev" or a zero length for an ordinary "write".

However that bug report refers to linux-image-2.6.24-16-xen so I spent a few hours last night looking through the diffs between 2.6.22-14 and 2.6.22-15 to see if there was a similar problem there. Well that part of the kernel has had a bit of an overhaul between 2.6.22 and 2.6.24, and I didn't come up with anything definitive.

I'll be looking at localedef when I get time. First up is running localedef under "strace" -- but I'm not going to get onto it for a while, so if someone else could do that and report what it's doing when it stops that would be really helpful.